Geologic maps serve as a valuable tool in resource exploration. Traditional mapping is characterized by time-intensive and high-cost fieldwork, but through multispectral and hyperspectral remotely sensed images it is possible to map extensive areas instantly.

Being able to calculate areas of interest for potential hydrothermal alteration zones can be extremely useful in determining the history and available resources geologically through remote sensing.

Bandpass Filters

Bandpass filters are essential for collection and analysis of satellite imagery, they allow specific ranges of electromagnetic wavelengths, or bands, to be transmitted while blocking others.

FWHM

Full Width Half Mass (FWHM) is an important concept for understanding how different bands are delineated.

The graph below visualizes the idea of FWHM. Each band in a spectral device only takes in a specific range of wavelengths, often the wavelength for each band is just denoted as the peak, but in reality each sensor is collecting a range of wavelengths. This is determined by denoting the FWHM as the area between the 50% transmittance on both sides of the curve.

Landsat 8, Hyperion, and ASTER Bands

In this plot you can see all the bands for Landsat 8, Hyperion, and ASTER. With their FWHM denoted as boxes. Its important to point out how Landsat 8 does have a band in the Short Wave Infrared (SWIR), but the FWHM is wide and this means that any SWIR data saved for this band could be attributed to any wavelength inside of the FWHM.

Location of Research

The Marysvale volcanic field in Utah is one of the largest volcanic fields in the western United States, known for its uranium and alunite deposits. The igneous rocks found here were intruded ~32-20 million years ago, due to the subduction of the Farallon plate, and extension of the Basin and Range.

Classification of Land Cover

It is important to eliminate areas that are not exposed soil or rock to accurately map alteration zones.

Land cover classification over North America is available for download from the National Land Cover Database. 2021 classification was used as that is the most recent year available.

The Landsat imagery tested against the NLCD was collected Oct 19th, 2023.

2021 NLCD was the most recent published dataset. I reduced the dimensions to three classifications: Barren, Snow, and Vegetation

Original RGB of Landsat 8 Image

Classification Tree

I selected 50 pixels of each class that I designated as Barren, Vegeations, and Snow. These were used in a CART (Classification and Regression Trees) algorithm.

Classified Landsat Image

Confusion Matrix

Using the NLCD as a reference model, I created a confusion matrix to determine the accuracy of the created predicted raster.

Class 1: Barren

Class 2: Snow

Class 3: Vegetation

## Confusion Matrix and Statistics
## 
##           Reference
## Prediction      1      2      3
##          1 253987     95 280759
##          2  10001     37    607
##          3  55484    176 399263
## 
## Overall Statistics
##                                          
##                Accuracy : 0.653          
##                  95% CI : (0.6521, 0.654)
##     No Information Rate : 0.6804         
##     P-Value [Acc > NIR] : 1              
##                                          
##                   Kappa : 0.3326         
##                                          
##  Mcnemar's Test P-Value : <2e-16         
## 
## Statistics by Class:
## 
##                      Class: 1  Class: 2 Class: 3
## Sensitivity            0.7950 1.201e-01   0.5866
## Specificity            0.5875 9.894e-01   0.8259
## Pos Pred Value         0.4749 3.476e-03   0.8776
## Neg Pred Value         0.8593 9.997e-01   0.4842
## Prevalence             0.3193 3.079e-04   0.6804
## Detection Rate         0.2539 3.698e-05   0.3991
## Detection Prevalence   0.5346 1.064e-02   0.4547
## Balanced Accuracy      0.6913 5.548e-01   0.7063

Sensitivity (True Positive Rate): The proportion of actual positives correctly predicted.

Specificity (True Negative Rate): The proportion of actual negatives correctly predicted.

Pos Pred Value (Positive Predictive Value): The proportion of predicted positives that are actually positive.

Neg Pred Value (Negative Predictive Value): The proportion of predicted negatives that are actually negative.

Prevalence: The proportion of actual instances of each class in the dataset.

Detection Rate: The proportion of actual positive instances correctly predicted.

Detection Prevalence: The proportion of predicted positive instances.

Balanced Accuracy: An average of sensitivity and specificity, useful when classes are imbalanced.

Masked Landsat Image

Using classification as a mask to keep only ‘barren’ pixels.

Mineral Group Mapping

Through specific combinations and equations of bands it is possible to create classification rasters for varying mineral groups and assemblages. To avoid over-classification I created a threshold which I deemed as only keeping the highest potential pixels.

Calcite Index

Calcite minerals can be mapped using (b6/b8) * (b9/b8)

Kaolinite, Sericite, Chlorite, and Epidote Index

Kaolinite, sericite, chlorite, and epidote minerals can be mapped using (b7+b9)/b8

Alteration Zones